A Comparative Assessment of Municipal Water Use in Turkey


Güneş M. Ş. , Yıldız D. , Kurnaz F. S.

JOURNAL OF WATER RESOURCES PLANNING AND MANAGEMENT, vol.148, no.2, 2022 (Journal Indexed in SCI) identifier

  • Publication Type: Article / Article
  • Volume: 148 Issue: 2
  • Publication Date: 2022
  • Doi Number: 10.1061/(asce)wr.1943-5452.0001514
  • Title of Journal : JOURNAL OF WATER RESOURCES PLANNING AND MANAGEMENT
  • Keywords: Sparse regression, Robust regression, Intrinsic variable selection, Missing data imputation, Municipal water assessment, SOLID-WASTE MANAGEMENT, CLIMATE-CHANGE, MISSING DATA, RANDOM FOREST, URBAN WATER, REGRESSION, IMPUTATION, ROBUST, RESOURCES, MODELS

Abstract

The use of assessment methods is essential for proper planning, improvement of the water system, and water economy in municipal water management. Thus, statistical model approaches can be used to evaluate the factors influencing water administration, understand their implications, forecast future use, and develop these systems accurately. The main purposes of this study are to reveal the current status of municipal water management in Turkey, assess the system with alternative regression models (with comparisons), and propose suggestions to managers as a decision support approach. Missing data imputation methods are practiced in order to improve the data set and model quality. According to the results, water extraction, income, and water treatment facilities are the most important issues in municipal water management. It is also seen from the modeling results that dams and treatment plants have a negative impact on municipal water use. In order to establish an adaptive municipal water management structure in Turkey, a number of management suggestions are proposed and assessed according to the models, which have intrinsic variable selection feature. Consequently, the least absolute shrinkage and selection operator (LASSO) may be practiced in short-term and cost-effective management planning. In long-term planning (with better model performance and high costs), elastic net and sparse least trimmed squares (LTS) may be preferred.